SF Autoresearch’s queue and the claimed edge for small experiments

A builder says SF Autoresearch prioritizes everyone’s first jobs, giving small experiments an advantage over large runs and keeping big buyers from buying up the whole platform. They also describe reinforcement-learning sandboxes that capture disk and memory state and can branch at any point to train agents on long tasks. They claim saving and restoring that state is fast enough to freeze a sandbox during tool calls, saving users about 7x.